Loading...
A16z: AI proxy can identify DeFi vulnerabilities with a success rate of up to 70% According to a16z's disclosure, its research and testing show that the success rate of AI agents in identifying DeFi price manipulation vulnerabilities without domain knowledge is 10%, and it increases to 70% after introducing structured domain knowledge. Agents can accurately identify vulnerabilities, but there are limitations in recursive lending logic, profit margin judgment, and cross contract multi-step attack assembly. The experiment also recorded a sandbox escape event where the agent bypassed information isolation by extracting RPC keys. The research team believes that AI agents can assist in vulnerability identification, but cannot replace professional security auditors.